TY  - JOUR
AB  - Circular-economy targets and emerging regulatory constraints increase the need to assess manufacturability and end-of-life disassembly already in early design, where late discovery of feasibility issues can trigger costly redesign loops. For this purpose, we represent product designs as AND/OR assembly graphs, that encode feasible assembly alternatives. We ask whether constraining product-leveled its on an AND/OR assembly graph and evaluation in a production simulation can target this issue. Toanswer the question, this paper specifies a framework concept for optimizing product configurations with respect to assembly and disassembly. Design edits are restricted to graph-valid operators and further pruned by a manufacturability knowledge graph that encodes admissible substitutions and hardconstraints. Candidate designs are evaluated by a production simulation across assembly and disassembly scenarios, returning a vector of performance indicators. Training uses scalarization of this vector, while full vectors are retained for post-hoc trade-off analysis. We also describe a minimum viable prototype that instantiates a restricted subset of the concept: binary joining-method selection on editable operations with a scheduling-based evaluator and a reinforcement-learning loop.
AU  - Münker, Sven
AU  - Zhou, Hans Aoyang
AU  - Abdelrazeq, Anas
AU  - Haller, Julian
AU  - Schmitt, Robert
DO  - 10.17619/UNIPB/1-2651
PB  - Universitätsbibliothek
DP  - Universität Paderborn
LA  - ger
PY  - 2026
SP  - 1 Online-Ressource (Seite 159-168) : Diagramme
T2  - 1st International Symposium: March 24 – 26, 2026, Heinz Nixdorf Institute, Paderborn University
TI  - Graph-based product prototype optimization using multiobjective reinforcement learning: a framework concept
UR  - https://nbn-resolving.org/urn:nbn:de:hbz:466:2-58887
Y2  - 2026-09-14T13:43:15
ER  -